The 95 Percent AI Pilot Failure Figure and What Its Source Supports

The chapter “Teams Misread Acceptance Evidence in Fourteen Predictable Ways” tells you to ask what was counted, who was counted, and how it was measured before you use a figure in a decision. This page applies that to one figure that keeps being re-quoted. It is here, not in the printed book, because the figure keeps being re-examined and the page can be updated.

The MIT NANDA report on generative AI in business, published in 2025, is often quoted as saying 95 percent of AI pilots fail. The report labels itself preliminary. It draws on a review of public initiatives, interviews with 52 organizations, and a survey of 153 leaders, and it says its figures are directionally accurate and come from interviews rather than company reporting. It states the 95 percent as a share of organizations in one place and as a share of custom tools reaching production in another (Challapally 2025). What it supports is that one preliminary, interview-based study reported few pilots reaching measurable business impact.

Last checked against the source: 2026-09-30. If you find a newer analysis or a correction, check it against the source before you rely on it.